PocketFlow

The core framework provides just enough structure a graph-based node execution model with shared state to implement sequential pipelines, branching logic, parallel execution, and multi-agent collaboration, while remaining small enough that the entire implementation can be understood and modified in a single reading.

The framework's central innovation is its approach to agent development: rather than having human engineers configure complex agent orchestration frameworks, PocketFlow provides the primitives through which an LLM can design and instantiate agent graphs to solve specific problems.

This 'agents building agents' paradigm reflects the observation that large language models are increasingly capable of reasoning about agentic system design, and that giving them a minimal, learnable framework produces better results than asking them to navigate framework abstractions built for humans.

PocketFlow is open-source and designed to be readable, forkable, and modifiable the 100-line constraint is intentional, ensuring that teams who adopt it can understand exactly what the framework does and customize it for their specific requirements without reverse-engineering complex dependency chains.

It is particularly useful for teams building AI agents who want full control over the execution logic rather than accepting the defaults and abstractions of larger frameworks like LangChain or LlamaIndex.